Drop in an image and get back the prompt that would produce it — one written the way Nano Banana likes it, one structured for GPT Image, and a short version for anything else. The page measures palette, lighting and framing itself; a vision model (your own key) or your own words fill in the subject.
These are computed from the pixels in this tab — no model involved. Click a swatch to copy its hex.
Model IDs go stale fast — if a call returns "model not found", pick Custom and type the current ID from the provider's model page.
The image goes directly from your browser to the provider you pick — it never touches mrg19.online. You're billed by that provider per call (usually a fraction of a cent for one image).
Palette, aspect ratio, lighting and detail level are filled in from the measurements in panel 1.